A new energy efficient data storage technology

Abstract

The application discloses a new energy efficient data storage technology and relates to the technical field of new energy data storage. The method comprises the following steps: before measured data collected by a new energy station reaches a storage system, obtaining predicted data of the new energy station, dividing the predicted data according to time windows and calculating a predicted change amount, comparing the predicted change amount with a judgment threshold, and marking each time window as a pre-judgment key period or a non-pre-judgment key period; extracting a collection time stamp of the measured data when the measured data is written, and searching for a pre-judgment mark of a corresponding period; if the period is a pre-judgment key period, a high-guarantee writing strategy is used to write the measured data into a storage medium, otherwise, a conventional writing strategy is used. The application pre-judges the data value by using the predicted data before the data is written, directly executes differentiated writing when the data is written, avoids indiscriminate consumption in the writing stage, and makes high-value data obtain differentiated storage guarantee as soon as the high-value data is generated.

Description

Technical Field

[0001] This invention relates to the field of new energy data storage technology, specifically to a method and system for predicting data value and implementing differentiated write control based on predicted data before writing measured data from new energy power plants into the storage medium. Background Technology

[0002] The new energy industry (wind power, photovoltaics, and energy storage) has experienced rapid development in recent years. Along with the increase in installed capacity, the amount of data generated by new energy power plants is rising exponentially. Taking wind farms as an example, sensor data on wind turbine vibration, temperature, and power are continuously generated at a frequency of seconds, with the annual raw time-series data volume of a single plant reaching terabytes. In energy storage power plants, battery management systems, energy management systems, and data acquisition and monitoring control systems continuously generate massive amounts of time-series data. This data needs to be collected, stored, and analyzed to support core business operations such as equipment status awareness, fault early warning, and operational optimization.

[0003] Currently, the mainstream architecture for new energy data storage is as follows: after the data is generated, it is all written to a time-series database or a distributed file system, and the storage space is subsequently managed through compression algorithms, hierarchical storage migration, or data eviction strategies.

[0004] In existing technologies, for example, CN119884138 discloses a method and system for managing equipment location data in an energy storage system, which writes data into caches of different importance levels by comparing the difference between data acquisition time and the current system time. Another example is CN121233057A, which discloses a data processing method, electronic equipment, and vehicle that performs multi-dimensional value assessment on stored memory data and determines a corresponding forgetting strategy based on the value assessment score.

[0005] While the aforementioned solutions optimize data storage from different perspectives, they all share a common structural flaw: a common characteristic of all existing solutions is that data writing occurs before value judgment. Data must be completely written to the storage system before the system can evaluate it; the write phase is "blind."

[0006] The direct consequences of the above-mentioned defects are: low-value data and high-value data occupy the same write bandwidth and storage resources when written; critical data cannot receive differentiated write treatment when it is generated and can only be compensated after post-analysis and confirmation; and the "store first, evaluate later, process later" model leads to additional input / output overhead and time costs. Summary of the Invention

[0007] To overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a new energy high-efficiency data storage technology. The problem to be solved is that the existing new energy data storage methods lack the ability to judge value when writing data to the storage medium. All data is written indiscriminately, resulting in the ineffective consumption of writing resources by low-value data, while high-value data cannot obtain differentiated storage protection when writing.

[0008] To achieve the above objectives, the present invention provides the following technical solution: Step 1: Obtain forecast data and perform time period prediction.

[0009] Before the measured data collected from the new energy power station reaches the storage system, the predicted data of the new energy power station is obtained, and the predicted data is a time series of predicted values.

[0010] The predicted data is at least one of the following: wind speed prediction from numerical weather forecast data of wind power plants; irradiance prediction from solar irradiance forecast data of photovoltaic power plants; and charge / discharge power setting values ​​output by the energy management system of energy storage power plants.

[0011] The predicted data is divided into multiple consecutive time windows according to a preset time window. The length of the time window is set according to the specific type of new energy power station: for wind power plants and photovoltaic power stations, the time window length is 5s to 10s; for energy storage power stations, the time window length is 1s to 3s.

[0012] For each time window, calculate the absolute value of the difference between the predicted value at the end of the time window and the predicted value at the beginning of the time window, and use this as the predicted change for that time window.

[0013] When the predicted data is a wind speed prediction or an irradiance prediction, the predicted change is the absolute value of the difference between the predicted value at the end of the time window and the predicted value at the beginning of the time window. When the predicted data is a charge / discharge power setting, the predicted change is the absolute value of the difference between the charge / discharge power setting at the end of the time window and the charge / discharge power setting at the beginning of the time window, or whether there is a charge / discharge state switching command within the time window.

[0014] When the predicted data is a set value for charging and discharging power, before comparing the predicted change in each time window with a preset judgment threshold, a state switching judgment step is also included: Obtain the charging / discharging state switching instructions output by the energy management system within each time window; When a charging / discharging state switching instruction is present within the time window, the time window is directly marked as a key prediction period, and the comparison step between the predicted change and the judgment threshold is no longer executed.

[0015] The predicted change amount for each time window is compared with a preset judgment threshold. When the predicted change amount is greater than or equal to the judgment threshold, the time window is marked as a key predicted period. When the predicted change amount is less than the judgment threshold, the time window is marked as a non-key predicted period, and a time period prediction marking result is generated.

[0016] The threshold for determination is the 80th percentile of the historical predicted change data of all time windows within a set time period for the new energy power station, where the set time period is 15 to 90 days.

[0017] Step 2: When writing the actual test data, query the prediction results for the corresponding time period.

[0018] When the measured data collected by the new energy power station arrives at the storage system and is ready to be written to the storage medium, the collection timestamp carried by the measured data is extracted.

[0019] Find the prediction mark corresponding to the time window to which the collection timestamp belongs in the predicted mark results for the time period.

[0020] Step 3: Perform differential write based on the query results.

[0021] When the detected prediction marker is a critical prediction period, a high-security write strategy is used to write the measured data to the storage medium; when the detected prediction marker is not a critical prediction period, a conventional write strategy is used to write the measured data to the storage medium.

[0022] The high-guarantee write strategy differs from the conventional write strategy in at least one dimension: write priority, number of replicas, index granularity, and storage media type.

[0023] Specifically, the difference between the high-guarantee write strategy and the regular write strategy in terms of write priority is as follows: the high-guarantee write strategy sends the test data into the write queue of the storage system with a higher write priority than the regular write strategy, so that the test data is written to the storage medium in the write queue with priority over the test data of the regular write strategy.

[0024] The difference between the high-guarantee write strategy and the regular write strategy in terms of the number of replicas is as follows: the high-guarantee write strategy creates 3 replicas of the test data in the storage system, while the regular write strategy creates 1 replica of the test data in the storage system.

[0025] The difference between the high-guarantee write strategy and the conventional write strategy in terms of index granularity is as follows: the high-guarantee write strategy establishes an independent index entry for each data point in the measured data, and the index entry at least includes the collection timestamp, data point number and value of the data point; the conventional write strategy establishes the index by minute, and all data points within the same minute share a single index entry.

[0026] The difference between the high-guarantee write strategy and the conventional write strategy in terms of storage media type is as follows: the high-guarantee write strategy writes the measured data to the solid-state drive storage area in the storage system, while the conventional write strategy writes the measured data to the mechanical hard drive storage area in the storage system.

[0027] This invention also provides a high-efficiency data storage system for new energy sources, comprising: The time period prediction module 101 is used to acquire the prediction data of the new energy power station, wherein the prediction data is a time series of the prediction values; the prediction data is divided into multiple consecutive time windows according to a preset time window. For each time window, calculate the absolute value of the difference between the predicted value at the end of the time window and the predicted value at the beginning of the time window, and use it as the predicted change for that time window. The predicted change amount for each time window is compared with a preset judgment threshold. When the predicted change amount is greater than or equal to the judgment threshold, the time window is marked as a key predicted period. When the predicted change amount is less than the judgment threshold, the time window is marked as a non-key predicted period, and a time period prediction marking result is generated.

[0028] The write execution module 104, whose input end is connected to the output end of the time period prediction module 101, is used to extract the collection timestamp carried by the measured data when the measured data collected by the new energy power station arrives at the storage system and is ready to be written to the storage medium, and to find the prediction mark corresponding to the time window to which the collection timestamp belongs in the time period prediction mark result. When the detected prediction marker is a key prediction period, a high-security write strategy is used to write the measured data into the storage medium. When the detected prediction marker is not a key prediction period, the measured data is written to the storage medium using a conventional writing strategy. The high-guarantee write strategy differs from the conventional write strategy in at least one dimension: write priority, number of replicas, index granularity, and storage media type.

[0029] The technical effects and advantages of this invention are as follows: This invention fundamentally avoids indiscriminate resource consumption during the write phase. By pre-judging the value before writing, low-value data is written using a conventional write strategy, without consuming the high-priority bandwidth, high-speed storage media, and multi-replica resources allocated by the high-guarantee write strategy; high-value data receives differentiated protection from the high-guarantee write strategy the moment it is generated.

[0030] This invention makes full use of the existing resources of renewable energy power plants. Predictive data is a standard data source for power prediction and scheduling management in renewable energy power plants. This invention does not require any additional hardware deployment; it can be implemented simply by deploying two software logic modules: a time period prediction module and a write execution module, resulting in low implementation costs.

[0031] The prediction logic of this invention is simple and computationally inexpensive. The core calculations of the time period prediction module are only subtraction operations on the predicted changes and comparison operations with the judgment threshold. It does not involve complex machine learning models or iterative calculations and will not have a noticeable impact on the write throughput of the storage system.

[0032] This invention has a wide range of applications and its parameters can be flexibly configured. The method of this invention can be applied to various new energy power plants such as wind power plants, photovoltaic power plants, and energy storage power plants. At the same time, key parameters such as the time window length and the judgment threshold can be flexibly set according to the historical data statistics of different power plants to meet diverse practical application needs. Attached Figure Description

[0033] Figure 1 This is a block diagram of the modular architecture of the system of the present invention; Figure 2 This is a flowchart illustrating the method of the present invention.

[0034] The attached diagrams are as follows: 101, Time Period Prediction Module; 102, Shared Storage Area; 103, Write Queue; 104, Write Execution Module. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Example 1 I. Definition of Terms The main terms involved in this invention are defined as follows for ease of understanding and implementation: Forecast data: In new energy power plants, data that can be obtained before the actual measured data is generated and which predicts the system's operating status at future times is called the time series of the forecast values.

[0037] For wind power plants, the forecast data is the wind speed forecast (in m / s) from the numerical weather prediction data. For photovoltaic power plants, the forecast data is the irradiance forecast (in W / m²) from the solar irradiance forecast data. 2 For energy storage power stations, the forecast data consists of the charge and discharge power setpoints (in kW or MW) and charge and discharge state switching instructions in the charge and discharge scheduling plan output by the energy management system.

[0038] Predictive data source: The system or device that generates the predicted data, including numerical weather prediction systems, light intensity prediction systems, and energy management systems.

[0039] Measured data: Data reflecting the actual operating status of the system, collected and uploaded in real time by the data acquisition terminal. This includes analog measurements such as voltage, current, power, temperature, vibration, speed, and irradiance, as well as digital signals such as circuit breaker position and equipment operating status. Measured data includes a timestamp of the acquisition.

[0040] Storage system: The target system used to receive and persistently store actual test data, including time-series databases and distributed file systems.

[0041] Time window: The basic time unit for dividing forecast data, with a start time and an end time.

[0042] Predicted change ( (This refers to the absolute value of the difference between the predicted value at the end of a time window and the predicted value at the beginning of a time window, used for wind farms and photovoltaic power plants. The unit is m / s or W / m².) 2 .

[0043] Power forecast change ( (This refers to the absolute value of the difference between the setpoint for charge / discharge power at the end of a time window and the setpoint for charge / discharge power at the beginning of a time window, used in energy storage power stations. The unit is kW or MW.)

[0044] Decision threshold (Th, ): Used to determine whether a time window falls within the critical threshold of a predicted key period. Th is used in wind farms and photovoltaic power plants. Used in energy storage power stations.

[0045] Predicting critical periods: By analyzing and forecasting data, we identify time windows where critical events may occur in the future. Critical events include sudden increases or decreases in power, rapid changes in wind speed, drastic fluctuations in irradiance, and transitions between charging and discharging states.

[0046] Non-predictable critical periods: Time windows determined by analyzing and forecasting data as where no critical events are expected to occur in the future.

[0047] Time period prediction marking results: The set of marking results generated after marking each time window is stored in the form of a time period prediction marking result table.

[0048] Shared storage area 102: A cache area in memory, which can be read and written by both the time period prediction module 101 and the write execution module 104, and is used to store the time period prediction marking results.

[0049] Write queue 103: A queue in the storage system used to cache data to be written. The actual data enters the queue in the form of data packets, and is retrieved by the write execution module 104 and written to the storage medium.

[0050] High-security write strategy: For actual test data during the predicted critical period, a write method with high write priority, multiple replicas, fine-grained indexes, and high-speed storage media is adopted.

[0051] Conventional write strategy: For actual test data not within the predicted critical period, the standard write priority, single copy, coarse-grained index, and ordinary storage medium are used for writing.

[0052] Write priority: The order in which actual data is scheduled to be written in write queue 103.

[0053] II. System Deployment Method The method of the present invention is deployed in the form of a software module in the local storage management system of the new energy power station, or in the edge data aggregation node of the power station.

[0054] Deployment of Time Period Prediction Module 101: The time-period prediction module 101 is deployed before the data access layer of the storage system, and is located on the same local area network as the prediction data source, acquiring prediction data through network communication. The time-period prediction module 101 can run as an independent process or as a background service of the storage system.

[0055] The time period prediction module 101 performs time period prediction according to the update cycle of the prediction data source. Each time the prediction data source completes a new prediction output, the time period prediction module 101 is triggered to perform time period prediction and writes the newly generated time period prediction mark result to the shared storage area 102, overwriting the old result from the previous one.

[0056] Deployment of execution module 104: The write execution module 104 is deployed in the write controller of the storage system and operates as a functional component of the write controller. The write execution module 104 is triggered when the measured data arrives at the write queue 103.

[0057] Preferably, the write execution module 104 is triggered by listening to the data enqueue event of the write queue 103, or by periodically polling the write queue 103 to check if new data has arrived.

[0058] Inter-module collaboration relationships: The time period prediction module 101 and the write execution module 104 operate independently. The time period prediction module 101 is triggered to execute when the predicted data source is updated, and the write execution module 104 is triggered to execute when the measured data arrives. The time period prediction module 101 pre-writes the time period prediction marking results into the shared storage area 102, and the write execution module 104 reads from the shared storage area 102 during the write operation. The two are decoupled through the shared storage area 102 and do not block each other.

[0059] The execution module 104 accesses the time period prediction mark result output by the time period prediction module 101 through shared memory, without increasing additional network communication overhead.

[0060] Resource consumption: The memory usage of the time period prediction module 101 depends on the time coverage and sampling density of the prediction data. Taking prediction data covering the next 4 hours with a sampling interval of 1 second as an example, the number of time windows that need to be stored is 2880 (taking a 5-second window as an example). The record size of each window is about 100B, and the total memory usage is about 0.3MB, which has a negligible impact on system resources.

[0061] III. Method for setting the judgment threshold The threshold Th (for wind farms and solar power plants) and The value of (energy storage power station) directly affects the number of critical periods predicted, which in turn affects the triggering frequency of high-security write strategies and the overall consumption of storage resources.

[0062] If the threshold is set too low, a large number of non-predicted critical periods will be mistakenly marked as predicted critical periods, causing the high-protection write strategy to be triggered too frequently and consuming too many storage resources. If the threshold is set too high, the truly critical periods will be missed, preventing high-value data from receiving differentiated protection during writes.

[0063] Therefore, the present invention provides a preferred method for setting the determination threshold, the specific steps of which are as follows: Step 1: Collect historical predicted change data for all time windows within a set time period for this renewable energy power station. The set time period is 15 to 90 days.

[0064] Step 2: Calculate the 80th percentile of the historical predicted change data, and use this value as the judgment threshold.

[0065] The 80th percentile means that, within a given time period, 80% of the predicted changes in any given time window are less than or equal to this value, and 20% of the predicted changes in any given time window are greater than this value. Using this value as a threshold means that approximately 20% of the periods with the most dramatic changes will be marked as critical periods for prediction.

[0066] The percentile value can be adjusted between 70 and 90 depending on the availability of actual storage resources. When storage resources are plentiful, a lower percentile can be used (to mark more time periods and provide higher data security), while when storage resources are scarce, a higher percentile can be used (to mark fewer time periods and save storage resources).

[0067] IV. Time Synchronization Requirements The method of this invention relies on the accurate matching of the timestamp of the measured data collection with the time window in the time period prediction mark result. Therefore, the data acquisition terminal of the new energy power station and the prediction data source connected to the time period prediction module 101 should use a unified time reference.

[0068] Specifically, the data acquisition terminal synchronizes with a standard time source via GPS or NTP protocol, with a synchronization accuracy better than ±100ms. When acquiring prediction data, the time period prediction module 101 records the time tag carried by the prediction data itself, which is also generated based on the standard time source.

[0069] Since both the data acquisition terminal and the prediction data source are traced back to the same standard time source (such as Coordinated Universal Time UTC), the time deviation between them is controlled within an acceptable range (not exceeding ±100ms). This deviation is much smaller than the minimum time window (1s) set in this invention, so it will not affect the accuracy of time window matching.

[0070] V. Application Examples (Wind Power Plants) Scene setting: A wind farm with an installed capacity of 100MW has 50 wind turbine generators. This wind farm is equipped with: The SCADA system's data acquisition terminal collects measured data such as voltage, current, power, speed, vibration, and temperature for each wind turbine, with a sampling frequency of 1 time per second. The data acquisition terminal maintains synchronization with a standard time source via the NTP protocol, achieving a time synchronization accuracy better than ±100ms.

[0071] The Numerical Weather Prediction (NWP) system outputs a second-by-second wind speed forecast for the next 4 hours every 15 minutes.

[0072] The time-series database is deployed on a local server as a storage system, equipped with a solid-state drive (500GB) storage area and a mechanical hard drive (10TB) storage area.

[0073] Time period prediction stage: Assuming the current time is 10:00:00 on July 29, 2026, the NWP system has just completed a new forecast output, providing second-by-second wind speed predictions for a total of 4 hours from 10:00:00 to 14:00:00.

[0074] The time period prediction module 101 obtains all wind speed prediction data for this forecast from the NWP system via RESTful API, totaling 14,400 data points. Each data point contains a timestamp and wind speed prediction value.

[0075] The time period prediction module 101 sets the time window length to 10 seconds, dividing the 14,400 data points into 1,440 consecutive time windows. The first window is from 10:00:00 to 10:00:10, the second is from 10:00:10 to 10:00:20, and the last is from 13:59:50 to 14:00:00.

[0076] The time-period prediction module 101 calculates the predicted wind speed change for each time window. Taking the window from 10:00:00 to 10:00:10 as an example: the predicted wind speed at the start is 7.2 m / s, and at the end is 7.8 m / s. The predicted change is... m / s.

[0077] The threshold Th for this wind field is set to 0.5 m / s / 10 s. The time period prediction module 101 will... m / s and Comparing m / s Therefore, this window is marked as a key time period for prediction.

[0078] Taking the window from 10:00:10 to 10:00:20 as an example: the wind speed changes from 7.8 m / s to 7.9 m / s. m / s, These are marked as non-predicted critical time periods.

[0079] The time-period prediction module 101 repeats the above process until all 1440 windows have been marked, and then writes the marking results to the shared storage area 102. Some results are shown below:

[0080] The time period prediction phase is complete; we are now awaiting the arrival of actual measurement data.

[0081] Write to the execution phase: At 10:00:03, the SCADA system data acquisition terminal collected a set of measured data, including approximately 500 data points of power, speed, vibration, etc. of 50 wind turbines at that time. The SCADA system packaged this set of data, carrying the acquisition timestamp 10:00:03, and pushed it to the write queue 103 of the time series database.

[0082] The write execution module 104 retrieves the data packet from the write queue 103 and extracts the collection timestamp. :00:03.

[0083] The write execution module 104 accesses the shared memory area 102, reads the time period prediction marking results, and searches for items containing... Within the time window, 10:00:03 belongs to the [10:00:00, 10:00:10] window, and the corresponding marker value is "critical".

[0084] Write to execution module 104 according to "Critical" - Select a high-security write strategy and perform the following operations: The test data is sent to the write queue 103 of the storage system with a higher write priority than the conventional write strategy. Create 3 copies in the storage system; Create an independent index entry for each data point; Write to the solid-state drive storage area.

[0085] After the write operation is complete, the time series database returns a write success response.

[0086] Subsequently, the measured data collected at 10:00:15 arrived. The write execution module 104 queried and found that the timestamp belonged to the [10:00:10, 10:00:20] window, with a mark value of "normal". The normal write strategy was selected. Write with standard priority; Create 1 copy; Indexes are built by minute; Write to the mechanical hard drive storage area.

[0087] Comparison results: In the above example, the data from 10:00:00 to 10:00:10 has high analytical value due to the rapid changes in wind speed, and a high-guarantee write strategy (high write priority, 3 replicas, fine-grained index, solid-state drive) is adopted during writing. However, the data from 10:00:10 to 10:00:20 has a more stable wind speed and lower analytical value, so a conventional write strategy (standard priority, 1 replica, coarse-grained index, hard disk drive) is used.

[0088] If the existing "write-before-evaluate" model is used, the data from the two time periods are completely indistinguishable during writing, requiring post-writing analysis and evaluation for differentiated processing. This invention performs differentiation processing during the writing process, avoiding the waste of writing resources and insufficient protection of high-value data.

[0089] VI. Parameter Configuration for Different Station Types For ease of implementation, the following table provides the optimal parameter configurations for various types of stations: 1. Wind power plant Forecast data source: Numerical weather forecasting system; Forecast data includes: wind speed (m / s); Time window length: 5s to 10s; Judgment threshold: 0.3 m / s to 0.8 m / s; Special criteria: None; High-security write strategy: 3 replicas, solid-state drives, point-by-point indexing, high priority; Standard write strategy: 1 replica, mechanical hard drive, minute-level index, standard priority; 2. Photovoltaic power station Predictive data source: Light intensity forecasting system; Predicted data content: Irradiance (W / m²) 2 ); Time window length: 5s to 10s; Judgment threshold: 50W / m 2 Up to 150W / m 2 ; Special criteria: None; High-security write strategy: 3 replicas, solid-state drives, point-by-point indexing, high priority; Standard write strategy: 1 replica, mechanical hard drive, minute-level index, standard priority; 3. Energy storage power station Forecast data source: Energy management system; Predicted data includes: charge / discharge power setting (kW / MW), and charge / discharge state switching commands; Time window length: 1 second to 3 seconds; Judgment threshold: 5% to 10% of rated power; Special judgment condition: When there is a charge / discharge state switching command, it is directly marked as a key prediction period; High-security write strategy: 3 replicas, solid-state drives, point-by-point indexing, high priority; Standard write strategy: 1 replica, mechanical hard drive, minute-level index, standard priority; Special criteria for determining energy storage power stations: When a charge / discharge state switching instruction is received within a time window, this time window is marked as a critical prediction period, regardless of whether the power prediction change exceeds the judgment threshold. This is because charge / discharge state switching is one of the most drastic operating condition changes in an energy storage power station, and the data generated during this period has extremely high analytical value, so it should be ensured that it is written with high security.

[0090] Description of composite scenarios: The above three scenarios are independently applicable. For new energy power plants that have multiple forecast data sources (such as integrated wind, solar and energy storage power plants), multiple forecast data can be used to predict time periods at the same time. When the prediction result of any forecast data is a critical time period, that time period is marked as the critical time period.

[0091] VII. Storage System Write Interface The implementation of this invention does not depend on a specific time-series database or distributed file system product. The time period prediction module 101 and the write execution module 104 interact with it through the standard write interface of the storage system.

[0092] The write execution module 104 reads the time period prediction tagging result from the shared storage area 102, obtains the Tag value corresponding to the current measured data, and then selects the corresponding write strategy based on the Tag value.

[0093] When the Tag is "critical", the write execution module 104 selects a high-security write strategy and passes the following parameter combination when calling the storage system write interface: write priority = high, number of replicas = 3, index granularity = point-by-point, storage medium = solid-state drive.

[0094] When the Tag is "normal", the write execution module 104 selects the normal write strategy and passes the following parameter combination when calling the storage system write interface: write priority = standard, number of replicas = 1, index granularity = minute level, storage medium = mechanical hard disk.

[0095] For time-series databases (such as InfluxDB and TimescaleDB), their write interfaces typically support specifying attributes such as write priority and replication factor through data point tags or write parameters. The write execution module 104 implements differentiated writes by passing different parameter values ​​in these interface calls.

[0096] For distributed file systems (such as HDFS), their write interfaces typically support differentiated writes by configuring different numbers of replicas and storage strategies.

[0097] Those skilled in the art can select the appropriate interface parameter configuration based on the specific storage product used in their site to implement the high-security write strategy and the conventional write strategy described in this invention.

[0098] VIII. Abnormal Handling The following abnormal situations may be encountered in the actual operation of this invention, and the corresponding handling methods are as follows: Exception 1: The prediction data source is temporarily unavailable.

[0099] When the time period prediction module 101 is unable to obtain prediction data (such as network interruption or prediction system failure), the time period prediction marking results in the shared storage area 102 may lack marking results for some future time periods.

[0100] Handling method: When the write execution module 104 cannot find the record corresponding to the collection timestamp of the current measured data, it sets the Tag to "normal" by default and executes the regular write strategy. At the same time, the time period prediction module 101 continuously attempts to reconnect to the prediction data source. Once the connection is restored, it immediately reacquires the prediction data and updates the time period prediction tagging result.

[0101] The reconnection and update of the tagging results are performed by the time period prediction module 101, while the write execution module 104 is only responsible for writing using the default strategy when no records are found. The two modules have separate responsibilities and do not interfere with each other.

[0102] This approach ensures that when the predictive data source temporarily fails, the storage system's write functionality remains unaffected, only temporarily losing its differentiated write capability.

[0103] Abnormal situation 2: There is a deviation between the timestamp of the measured data collection and the standard time source.

[0104] When there is a time synchronization deviation in the data acquisition terminal, the timestamp of the measured data acquisition may not match the time window in the time period prediction marking result.

[0105] Solution: Configure GPS or NTP time synchronization service when deploying the data acquisition terminal, with a synchronization interval of no more than 1 hour, ensuring that the time deviation is controlled within ±100ms. Before deploying this invention, the time synchronization configuration of the data acquisition terminal should be checked and calibrated.

[0106] Abnormal situation 3: Access to shared storage area 102 is abnormal.

[0107] When the time period prediction module 101 or the write execution module 104 cannot read or write the shared memory area 102 normally (e.g., the shared memory is full or the read / write lock times out), the following fallback processing is executed: When the time period prediction module 101 is unable to write to the shared storage area 102, it temporarily stores the prediction mark result of the current time period in the local cache, and rewrites it after the shared storage area 102 is restored to access.

[0108] When the write execution module 104 is unable to read the shared storage area 102, it sets the Tag of the current measured data to "normal" by default and executes the normal write strategy to ensure that the write function is not affected.

[0109] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A new energy-efficient data storage technology, characterized in that, Includes the following steps: Before the measured data collected from the new energy power station reaches the storage system, the predicted data of the new energy power station is obtained, and the predicted data is a time series of predicted values; The predicted data is divided into multiple consecutive time windows according to a preset time window. For each time window, calculate the absolute value of the difference between the predicted value at the end of the time window and the predicted value at the beginning of the time window, and use it as the predicted change for that time window. The predicted change amount for each time window is compared with a preset judgment threshold. When the predicted change amount is greater than or equal to the judgment threshold, the time window is marked as a key predicted period. When the predicted change amount is less than the judgment threshold, the time window is marked as a non-key predicted period, and a time period prediction marking result is generated. When the measured data collected by the new energy power station arrives at the storage system and is ready to be written to the storage medium, the collection timestamp carried by the measured data is extracted. Search the predicted marker corresponding to the time window to which the collection timestamp belongs in the predicted time period marker results; When the detected prediction marker is a critical prediction period, a high-security write strategy is used to write the measured data to the storage medium; when the detected prediction marker is not a critical prediction period, a conventional write strategy is used to write the measured data to the storage medium. The high-guarantee write strategy differs from the conventional write strategy in at least one dimension: write priority, number of replicas, index granularity, and storage media type.

2. The new energy high-efficiency data storage technology according to claim 1, characterized in that, The difference between the high-guarantee write strategy and the regular write strategy in terms of write priority is as follows: The high-security write strategy sends the test data into the write queue of the storage system with a higher write priority than the conventional write strategy, so that the test data is written to the storage medium in the write queue with priority over the test data of the conventional write strategy.

3. The new energy high-efficiency data storage technology according to claim 1, characterized in that, The difference between the high-guarantee write strategy and the conventional write strategy in terms of the number of replicas is as follows: The high-guarantee write strategy creates three copies of the test data in the storage system, while the regular write strategy creates one copy of the test data in the storage system.

4. The new energy high-efficiency data storage technology according to claim 1, characterized in that, The difference between the high-guarantee write strategy and the conventional write strategy in terms of index granularity is as follows: The high-security write strategy establishes an independent index entry for each data point in the measured data. The index entry includes at least the collection timestamp, data point number, and value of the data point. The conventional write strategy creates an index every minute, with all data points within the same minute sharing a single index entry.

5. The new energy high-efficiency data storage technology according to claim 1, characterized in that, The difference between the high-security write strategy and the conventional write strategy in terms of storage media type is as follows: The high-security write strategy writes the measured data to the solid-state drive storage area in the storage system, while the conventional write strategy writes the measured data to the mechanical hard drive storage area in the storage system.

6. The new energy high-efficiency data storage technology according to claim 1, characterized in that, The predicted data is at least one of the following: Wind speed forecasts from numerical weather prediction data for wind farms; The predicted irradiance value in the solar intensity forecast data of photovoltaic power plants; The setpoint for charging and discharging power output by the energy management system of the energy storage power station.

7. The new energy high-efficiency data storage technology according to claim 6, characterized in that: When the predicted data is a wind speed prediction value, the length of the time window is 5s to 10s, and the judgment threshold is 0.3m / s to 0.8m / s; When the predicted data is a predicted irradiance value, the length of the time window is 5 to 10 seconds, and the judgment threshold is 50 W / m². 2 Up to 150W / m 2 ; When the predicted data is a set value for charging and discharging power, the length of the time window is 1 to 3 seconds, and the judgment threshold is 5% to 10% of the rated power of the energy storage power station.

8. The new energy high-efficiency data storage technology according to claim 1, characterized in that, The determination threshold is set according to the following steps: Collect historical predicted change data for all time windows within a set time period, which is 15 to 90 days, for the new energy power station. Calculate the 80th percentile of the historical predicted change data, and use the value of the 80th percentile as the judgment threshold.

9. The new energy high-efficiency data storage technology according to claim 1, characterized in that, When the predicted data is a set value for charging and discharging power, before comparing the predicted change in each time window with a preset judgment threshold, a state switching judgment step is also included: Obtain the charging / discharging state switching instructions output by the energy management system within each time window; When a charging / discharging state switching instruction is present within the time window, the time window is directly marked as a key prediction period, and the comparison step between the predicted change and the judgment threshold is no longer executed.

10. A high-efficiency data storage system for new energy sources, characterized in that, include: The time period prediction module (101) is used to acquire the prediction data of the new energy power station, wherein the prediction data is a time series of the prediction value; the prediction data is divided into multiple consecutive time windows according to a preset time window. For each time window, calculate the absolute value of the difference between the predicted value at the end of the time window and the predicted value at the beginning of the time window, and use it as the predicted change for that time window. The predicted change amount for each time window is compared with a preset judgment threshold. When the predicted change amount is greater than or equal to the judgment threshold, the time window is marked as a key predicted period. When the predicted change amount is less than the judgment threshold, the time window is marked as a non-key predicted period, and a time period prediction marking result is generated. The write execution module (104) has its input end connected to the output end of the time period prediction module (101). When the measured data collected by the new energy power station arrives at the storage system and is ready to be written to the storage medium, it extracts the collection timestamp carried by the measured data and searches for the prediction mark corresponding to the time window to which the collection timestamp belongs in the time period prediction mark result. When the detected prediction marker is a key prediction period, a high-security write strategy is used to write the measured data into the storage medium. When the detected prediction marker is not a key prediction period, the measured data is written to the storage medium using a conventional writing strategy. The high-guarantee write strategy differs from the conventional write strategy in at least one dimension: write priority, number of replicas, index granularity, and storage media type.

Images